Cell Type Based Cell-Cell Communication Pattern And Application

In multicellular organisms,cells communicate with each other through various signaling factors.Cell-cell communication is essential for the development of organisms and maintenance of health.Specifically,cell-cell communication regulates cell differentiation,organ development,tissue repair,and pathogen clearance.Furthermore,cell-cell communication plays a crucial role in the occurrence and progression of diseases.For instance,in the context of cancer,cell-cell communication can help the immune system recognize,but it can also help tumor cells escape the pursuit of immune cells.Therefore,studying cell-cell communication contributes to understanding of cell development,tissue homeostasis,and guides the diagnosis and treatment of diseases.With the maturity of singlecell RNA sequencing(scRNA-seq)technology,it provides cell-level resolution for cell-cell communication research,identifying "which cells communicate with each other" and offering a gene-level perspective to infer "which signaling molecules are involved in communication."This article is based on the perspective of cell types and their hierarchical structure,using single-cell transcriptome sequencing data to explore and analyze cell-cell communication patterns at both the cell and gene levels.At the cell level,a single-cell clustering framework integrating pathway data is constructed,which can more accurately and robustly identify cell type labels required for cell-cell communication analysis.Furthermore,within interacting cell types,cell sub-crosstalk pairs are defined and identified,providing a finer-grained perspective for cell-cell communication analysis.At the gene level,a robust evaluation framework for cell-cell communication inference methods is designed to systematically evaluate the robustness of cell-cell communication inference methods and provide userfriendly tools.In addition,we inferred pan-cancer communication patterns in different cell hierarchical structures,and then analyzed the common and specific pan-cancer communication patterns between different hierarchical structures.The main research contents and contributions of this article are as follows:We construct a single-cell clustering framework integrating pathway data called sci Path(single-cell clustering integrating pathway),which can effectively reduce the impact of noise on single cell clustering methods and improve their accuracy and robustness.This framework initially fuses single-cell transcriptomic data with pathway data using the similarity network fusion(SNF)algorithm to present fused cell similarity.It then employs the cell similarity matrix as input to existing single-cell clustering methods.Using sci Path,we integrate 26 scRNA-seq datasets and 4 pathway databases into 10 state-of-art single-cell clustering methods.We evaluate the clustering results using three accuracy metrics,three noise simulation strategies,and robustness metrics.The evaluation results show that integrated pathway data effectively enhances the performance of single-cell clustering methods,with the best performance observed when integrating de novo pathways into the SC3 single-cell clustering method.Furthermore,by studying the impact of pathway redundancy on performance improvement,we find that reducing highly redundant pathway data(e.g.,Reactome)can enhance the clustering performance of integrated pathway data.We define cell sub-crosstalk pairs(CSCP)to characterize fine-grained cell communication patterns.Utilizing scRNA-seq data,cell type labels,and ligand-receptor data,we define CSCP as combinations of sender cell subpopulations and receiver cell subpopulations with similar(low expression difference score)and strong(high communication score)communication signals.Two factors are taken into account in the definition: the communication signal similarity within cell subpopulations based on expression differences and communication signal strength between cell subpopulations.Using spatial transcriptomic data to calculate cell distances within CSCP,we validate that CSCP effectively characterizes local cell communication patterns.Furthermore,we use 13 cell-cell communication inference methods to infer ligand-receptor pairs within CSCP,demonstrating that CSCP reveals communication patterns masked at the cell type level.Additionally,applying CSCP to predict the anti-PD-1 therapy prognosis of 29 breast cancer patients confirms that CSCP is an effective feature to predict immunotherapy outcomes.We establish a robustness evaluation framework for cell-cell communication inference methods and develop a user-friendly tool called robust CCC.This framework includes 14 well-recognized cell-cell communication inference methods and 6 noise simulation strategies.We use robust CCC to evaluate the robustness performance of all included cellcell communication inference methods on 48 single-cell RNA sequencing datasets from the mouse cerebellar cortex.The results indicate that these methods exhibit distinct robustness performance in different types of noisy data.Furthermore,network-based methods do not outperform other methods in robustness,but Cyto Talk,due to its strategy of reconstructing gene regulatory networks,outperforms other network-based methods.Finally,we emphasize that methods with low robustness do not necessarily indicate poor accuracy,it may be due to sensitivity to data variations,and this sensitivity is an advantage of the method,allowing the detection of weak cell-cell communication signals.We define pan-cancer communication patterns at different cell hierarchical levels and compare the commonalities and differences in pan-cancer communication patterns among different levels.Based on the Tumor Immune Single Cell Hub 2(TISCH2),which contains single-cell transcriptomic data for 15 cancers,as well as cell annotations for cell types,cell subtypes,and secondary cell subtypes,we use the ligand-receptor analysis framework(LIANA)to identify ligand-receptor pairs between tumor cells and immune cells in each hierarchical level.We then define and identify pan-cancer communication patterns based on statistical significance.Calculating the similarity of pan-cancer communication patterns among different levels,we find substantial differences.Further functional enrichment analysis using GO,KEGG,and Reactome database suggests that more immune-related functions are enriched in fine-grained hierarchical structures,which are not enriched in coarse-grained hierarchical structures.Moreover,based on expression profiles and clinical data from The Cancer Genome Atlas(TCGA)for 15 cancers,we perform survival analysis of pan-cancer communication patterns,revealing a positive correlation between communication scores and survival differences.As the hierarchical granularity increases,this correlation becomes more pronounced.Finally,we analyze the significance of immune checkpoint communication scores at different hierarchical levels,revealing that communication scores become more significant as hierarchical granularity increases.